1. The Nature of the Driving Force: From Consumer Electronics Inventory Pulls to AI Computing Power Expansion

The core of the previous substrate boom (2019–2022) was a massive inventory pull driven by surging PC and laptop demand during the pandemic. The entire industry expanded aggressively under client commitments. However, when the pandemic dividend ended and consumer demand halved, the painful inventory corrections and price collapses of 2023 and 2024 ensued.

This time, the demand structure is fundamentally different:

• Explosive Growth in Substrate Usage for AI GPUs and ASIC Accelerators: From the Hopper generation to the Blackwell generation, the ABF substrate area required for a single GPU has grown by more than double; moving from Blackwell to Rubin, it will double again. This is not simply an "increase in volume," but a structural demand shift where "unit consumption area, layer count, and process difficulty are leaping simultaneously."
• General Server Upgrades Contributing to 20–50% Content Growth: As CPUs undergo generational shifts, their physical sizes increase and layer counts rise. For substrate suppliers, this represents an invisible yet tangible revenue boost.
• 800G/1.6T Switch Generational Shifts Driving Double Consumption: As switches transition from 400G to 800G to 1.6T, each generational leap roughly doubles the per‑unit substrate consumption, creating strong demand for high‑layer‑count ABF substrates used in networking.
• Agentic AI Pushing CPU Demand: Every AI workload requires roughly 3 to 5 times the CPU core count compared to traditional workloads. This implies that the server CPU market size could expand by 4 to 5 times over the next five years.

In short, this is not a "short‑term inventory pull" for consumer electronics, but a "structural, long‑cycle investment" in cloud infrastructure and AI computing power build‑outs.

2. The Bottleneck Material: Supply Constraints of T‑glass Fibreglass Cloth

Key Takeaway: If you only remember one point from this article, it should be this – the core bottleneck of the current substrate shortage lies in upstream materials, not the substrate production capacity itself. Specifically, Low CTE (Coefficient of Thermal Expansion) fibreglass cloth, commonly known as T‑glass, is the critical material for high‑frequency, high‑speed, multi‑layer substrates. Its supply status dictates the ceiling of the entire ABF substrate supply chain.

The True Landscape of the T‑glass Supply Structure

• Nittobo alone controls over 90% of the global market. Most of its output goes to major CCL manufacturers like Resonac, which then flows to large‑format FCBGA substrate plants like Ibiden and Unimicron.
• Nittobo has announced an investment of approximately ¥15 billion to expand its Fukushima plant, aiming to triple T‑glass capacity. The new production line is expected to commence operations in 2027 (fiscal year 2026 Q4). In addition, Nittobo plans to launch a next‑generation T‑glass in 2028, reducing the CTE from 2.8 ppm to approximately 2.0 ppm.
• Secondary suppliers such as Baotek, Taiwan Glass, Nan Ya Plastics, Taishan Fibreglass, and Fulltech Fibreglass are aggressively expanding capacity and seeking certifications, but their yields and actual shipment volumes remain on the learning curve.

Practical Implications for the Downstream

In the short term, the T‑glass shortage will not be resolved. Even if all fibreglass cloth manufacturers' expansions land on schedule, this critical material will remain tightly supplied until 2027 or even 2028. This means:

• Upstream material price hikes will continue to pass downstream; substrate quotes still have room to rise.
• To secure materials, clients are willing to make prepayments, sign LTAs, or even make strategic equity investments.
• Tier‑2 substrate makers without material guarantees will be squeezed out of high‑end orders.

3. Differences in Expansion Strategies Among Taiwan's Top Three Substrate Manufacturers

Regarding the expansion plans of Taiwan's three major substrate makers – Unimicron, Kinsus, and Nan Ya PCB – it is worth understanding their strategic differences in detail. | Manufacturer | Core Strategy | Capacity & Expansion Plans | Key Highlights | | :--- | :--- | :--- | :--- | | Unimicron (欣興) | Comprehensive Technical Leadership | ABF capacity planned to grow ~40% YoY in 2026. High‑end capacity concentrated in Guangfu (Phase 3 ramping up H2 2026, Phase 4 mass production 2027) and Yangmei plants. | AI‑related revenue share expected to rise from ~40% to 60%. Converting BT lines to ABF, showing high confidence in structural demand. HDI yields improving, volume scaling from Q2 2026. | | Kinsus (景碩) | Betting on CPU Substrates & New AI Projects | Expansion budget of ~NT$23.5 billion for 2026–2028, largely focused on the K6 plant. 2026 CAPEX spiking to NT$9.5B (from NT$5.5B in 2025). | ABF monthly capacity expanding from ~40M to 50M units (K6B mass production in early 2027). Significant opportunity to capture market share in NVIDIA Vera CPU substrates (growing from ~10% in the Grace generation to ~50%). | | Nan Ya PCB (南電) | Strong Pricing Flexibility, Conservative Expansion | Large‑scale expansion won't significantly contribute until 2028. Kunshan plant utilising debottlenecking in 2026 for double‑digit % capacity growth. | Higher proportion of spot orders allows them to enjoy price‑hike dividends faster than peers. Backed by the Formosa Plastics Group, granting inherent advantages in procuring critical materials like fibreglass cloth and copper foil. |

4. The Pricing Structure is Reshuffling: Why LTAs are No Longer Tools for Depressing Prices

Many people associate "Long‑Term Agreements" (LTAs) with "price locking," assuming that signing an LTA means sacrificing future upside pricing. However, the dynamics of LTAs in this cycle are completely different:

• LTAs lock in volume, not price: Most long‑term agreements currently being signed retain clauses allowing prices to float with material costs.
• Clients are willing to pay upfront deposits: Cloud Service Providers (CSPs) and major chip makers are aligning prepayments with expansion schedules just to secure capacity.
• Strategic investments (private placements) are emerging: Substrate manufacturers are accepting client investments to lock in long‑term order visibility and diversify expansion risks.
• In May 2026, Ajinomoto notified clients of a 30% price hike for ABF build‑up film, effective Q3. This cost pass‑through is bound to continue.

In practice, this means:

ABF substrate quotes are projected to see a 15–20% YoY increase for the full year of 2026, with price hikes potentially accelerating further in 2027. For downstream PCB engineering integrators, IC packaging houses, and end‑market OEM/ODMs, pricing strategies, material management, and supplier negotiation rhythms must be entirely recalibrated.

5. The Battle of Three Next‑Generation Packaging Technology Routes

Aside from the current explosion in ABF substrate demand, the evolutionary path of next‑generation packaging technologies requires close monitoring. Three technical routes are currently competing, each bringing different impacts to the substrate supply chain.

Route 1: Glass Substrates

• Advantages: Superior dimensional stability, heat dissipation, and electrical performance compared to organic substrates. TGV (Through‑Glass Via) technology allows for smaller via diameters and higher‑density interconnections.
• Challenges: Hole consistency and surface roughness in glass drilling remain hurdles. Multi‑layer alignment and via‑filling reliability are yield bottlenecks.
• Timeline: Limited mass production will only appear in 2027–2028. Ibiden expects broad adoption not until 2030; pilot lines from SEMCO and Sumitomo Chemical won't enter mass production until after 2027.

Route 2: CoWoP (Chip‑on‑Wafer‑on‑PCB)

• Concept: Led by NVIDIA in collaboration with TSMC, this involves bonding the silicon interposer along with the GPU/HBM directly onto a large‑format PCB, eliminating the ABF substrate layer.
• Benefits: Expected to reduce packaging costs by 30–50% and support panel‑level manufacturing (e.g., 450×450 mm).
• Challenges: Requires compressing PCB line resolution from the current 20/35 μm down to under 10 μm, far exceeding current PCB process capabilities. Issues with warpage, CTE mismatch, and reliability remain unsolved.
• Industry Significance: A potential threat to substrate makers, but an opportunity for high‑end HDI/mSAP PCB manufacturers.

Route 3: EMIB‑T (Embedded Multi‑die Interconnect Bridge with TSV)

• Concept: Led by Intel, this involves embedding small silicon bridges into organic ABF substrates and adding TSVs to enhance power delivery and signal transmission.
• Timeline: Intel targets mass production between H2 2027 and 2028, with potential applications in Google TPU v9, Meta MTIA, and MediaTek's next‑gen ASICs.
• Industry Significance: A stabilising net‑positive for the ABF ecosystem. It doesn't replace organic substrates but enhances their packaging functionality. It demands finer line widths/spacing, more precise laser drilling, and cavity‑embedding processes from substrate makers. In short: In the short to medium term, the position of ABF substrates remains rock solid. Glass substrates and CoWoP are still in the technical validation phase. For the Taiwanese supply chain, the most critical immediate focus remains yield improvement and material support for high‑end ABF substrates.

6. Practical Implications for the Taiwanese Electronics Supply Chain

Based on the above observations, we have compiled several actionable directions for Taiwanese PCB/substrate professionals, as well as associated R&D, procurement, quality, and sales teams:

1. Pricing Strategies Require Dynamic Adjustments: Upstream material price hikes will not stop passing down. Quoting new projects using the cost structures of the past six months will erode gross margins. It is recommended to establish monthly material cost tracking sheets to sync with client quoting cycles.
2. Material Management Takes Priority Over Price Management: The supply of critical materials like T‑glass, ABF films, and specific copper foils may impact delivery commitments more than price negotiations. Rather than squeezing suppliers on price, focus on securing material allocations.
3. The Shift Towards High‑End Client Structures is Irreversible: Growth in consumer electronics and standard HDI is limited. The real growth engines are AI servers, networking equipment, and high‑layer‑count general servers. R&D resources, production line allocation, and business development should aggressively pivot in this direction.
4. Keep an Eye on Next‑Gen Packaging, But Don't Over‑Anxiety: Glass substrates and CoWoP still have a time window of 2–3+ years. The real problems to solve today are the yields, layer counts, and fine‑line capabilities of high‑end ABF substrates.
5. The Time to Build Deep Partnerships with Downstream Clients is Now: When clients are willing to prepay, invest equity, and sign LTAs, it creates a rare bargaining window not seen in a decade. For mid‑sized PCB/substrate makers, this is a critical moment to establish long‑term trust.

Conclusion: Standing at the Beginning of a Multi‑Year Economic Cycle, Not the End

This substrate upcycle is fundamentally different from the past PC inventory pulls or mobile phone specification upgrades. It is not a one‑to‑two‑year inventory restocking event, but a multi‑year expansion of AI computing infrastructure.

From the T‑glass supply bottlenecks to the differentiated expansion strategies of Taiwan's big three manufacturers, to the reshuffling of pricing structures, and the battle of next‑generation packaging technologies – every link points to the same conclusion:

The substrate industry has entered a new phase characterised by a different structure, longer cycles, and higher technical barriers.

For professionals in the Taiwanese electronics supply chain, this is both a challenge and a rare opportunity. The challenge lies in the simultaneous pressures of cost pass‑throughs, material management, and technological upgrades. The opportunity is that those who can perfect their process capabilities, material support, and customer service will firmly secure their positions in the high‑end supply chain during this cycle.